#3434th of 39 in RAG, knowledge and search
deep-searcher
Python library for multi-step research reports over private documents in Milvus
- Stars
- 8.4k
- License
- Apache-2.0
- Last commit
- Sep 2026
- Language
- Python
7,641 on 1 Apr 20268,350 on 21 Oct 2026
From GitHub's daily star counts.
Checked on GitHub.
Zilliz is hiring: 11 open roles
Overview
DeepSearcher is a Python package that loads local files (and web pages via Firecrawl) into a Milvus vector database, then answers questions by running iterative search and reasoning steps with an LLM to produce a written report. LLM, embedding model and vector store are chosen through set_provider_config calls, with many hosted providers and Ollama supported.
Who it is for: Python developers building private-data research and Q&A on Milvus
Strengths
- Many LLM providers: OpenAI, DeepSeek, Anthropic, Gemini, Grok, Bedrock, watsonx, Ollama
- Local embedding options via Ollama, FastEmbed and Pymilvus models
- Installs with pip; optional extras such as deepsearcher[ollama]
- Produces long-form reports rather than single-shot answers
Weaknesses
- Milvus is the only vector database; other backends are marked under development
- Web crawling needs a Firecrawl API key and is described as under development
- Quick start defaults to OpenAI and requires OPENAI_API_KEY
- Python API only; no bundled web UI documented in the visible README
What it needs
- GPU optional
- Docker
- Needs Milvus or Zilliz Cloud, LLM API key or local Ollama, Firecrawl (optional, web crawling)
- Models: OpenAI, DeepSeek, Anthropic, Gemini, xAI Grok
Also in RAG, knowledge and search
See all 39| Rank | Project | Score |
|---|---|---|
| 1 | deeptutorSelf-hosted AI tutor with knowledge bases, guided learning, books and quizzes | 82 out of 100 |
| 2 | LightRAGGraph-based RAG framework with a REST API server and web UI | 78 out of 100 |
| 3 | MorphicAI answer engine with cited search results and generative UI components | 77 out of 100 |
| #4 | MaxKBEnterprise knowledge-base agent platform with RAG, workflows and MCP tools | 75 out of 100 |
| #5 | RAGFlowRAG engine with deep document parsing, template chunking and agentic retrieval | 74 out of 100 |